I work in AI and computer vision, which means I spend my days teaching machines to make sense of the visual world. Then I go home and try to do the same with my own life, with significantly worse results.

I’m based in Brașov, Romania. I’m Head of Artificial Intelligence at Dynamic Ventures Inc. (d/b/a CountThings), where the work splits roughly in two: training computer vision models — deep networks built on human-labeled data, because that’s what reliably solves real problems at scale — and working with the people who build them. The second half turns out to be the harder problem, and the one I’ve learned more from.

But supervised learning isn’t the question I find most interesting.

What keeps pulling me back is a different one: how far can systems get without all those labels? Systems that learn mostly by observing — by predicting what’s missing, or what comes next — with explicit feedback arriving late and sparingly, roughly the way humans and animals appear to learn. This isn’t my idea. It’s an active research direction (self-supervised learning, predictive objectives, world models), and it’s an open question, not a settled result. I think it’s the most plausible bet currently on the table. I might be wrong.

AVM Innovation is where I write that bet down. It’s a manifesto, not a company — no products, no revenue, nothing to sell. It’s the lens I bring to my work at Dynamic Ventures and to the projects I’m part of: a way of designing software and AI that stops treating data as ground truth and starts treating it as what it actually is — a partial, instrument-shaped projection of a world we can only make sense of by learning to anticipate its gaps. It may turn into something more one day. Right now it’s a position, held in public.

I also help with Povești Care Cresc and its English counterpart, Ava’s Tales Garden — a children’s book project started by my closest friend. The idea and the work are his; I contribute where I’m useful, and there may eventually be some apps that I’d build. I’m involved because I think critical thinking and thoughtful education are the quiet foundations of adaptable generations, and because books that take children seriously are rarer than they should be.

I was involved for a while with Asociația În Numele lui Alexandru, a non-profit founded by the parents of my childhood friend Alexandru, whom we lost in the Colectiv fire in Bucharest. I’m not active there these days, but I help when I’m asked and when I can. In small and imperfect ways, I try to support projects that push for a more responsible, less indifferent society.

I’ve always been drawn to what curiosity does when you let it cross borders. The same patterns keep surfacing in unrelated fields — physicists, psychologists, biologists and machine learning engineers rediscovering each other’s concepts and renaming them, apparently without talking to each other. I find that both fascinating and slightly ridiculous. Lenses is where I try to map those overlaps — not as an expert, but as someone who reads across fields and can’t stop noticing them. Worth saying plainly: an analogy that holds between two fields is a starting point, not evidence. Sometimes the same shape means the same mechanism. Usually it doesn’t.

I’m not an academic. I’m not a guru. I’m an engineer who reads too much, connects things that probably shouldn’t be connected, and has a persistent suspicion that the universe runs on fewer subroutines than it wants us to believe. I treat most explanations with the kind of polite skepticism that tends to annoy people who are very sure of themselves — including, on good days, myself.

A few things worth knowing before you go further:

• I don’t offer answers. I offer perspectives — a small, arbitrary subset of the infinite number one could construct on any topic. I’m aware of this. I’ve simply chosen not to write all of them, partly for practical reasons, partly out of respect for the finite human lifespan.
• Every piece is versioned. It’s a snapshot of how I understand something at the time of writing, which means it will change, get contradicted by something I read next Tuesday, or quietly embarrass me in three years. That’s fine. The alternative is pretending I’ve figured it all out, and we both know how that usually ends.
• I grew up Orthodox and no longer believe most of it. I still say “God help me” when things get difficult, which is either hypocritical or human — I’ve decided it’s probably both. Most of my conclusions end this way.

I write as a way of thinking in public. If something here resonates, or you see a perspective I missed — and you will — I’d like to hear it.